RACE is a large-scale reading comprehension dataset with more than 28,000 passages and nearly 100,000 questions. The dataset is collected from English examinations in China, which are designed for middle school and high school students.
Dataset Card
RACE is a large-scale reading comprehension dataset with more than 28,000 passages and nearly 100,000 questions. The dataset is collected from English examinations in China, which are designed for middle school and high school students. The dataset can be served as the training and test sets for machine comprehension. An example of 'train' looks as follows. An example of 'train' looks as follows. An example of 'train' looks as follows. The data fields are the same among all splits. - exampleid: a string feature. - article: a string feature. - answer: a string feature. - question: a string feature. - options: a list of string features. - exampleid: a string feature. - article: a string…
Excerpt from the card by Eduard Hovy, licensed other.
Structure
all 97,687 rows
| Split | Rows | Size |
|---|---|---|
| test | 4,934 | 9.8 MB |
| train | 87,866 | 175.3 MB |
| validation | 4,887 | 9.7 MB |
high 69,394 rows
| Split | Rows | Size |
|---|---|---|
| test | 3,498 | 7.0 MB |
| train | 62,445 | 124.6 MB |
| validation | 3,451 | 6.9 MB |
middle 28,293 rows
| Split | Rows | Size |
|---|---|---|
| test | 1,436 | 1.8 MB |
| train | 25,421 | 31.0 MB |
| validation | 1,436 | 1.8 MB |
Details
- Repository
- ehovy/race
- Publisher
- Eduard Hovy
- Task category
- Multiple choice
- Tags
- Not stated by the source
- Size category
- 10K<n<100K
- Languages
- en
- Revision
- 2fec9fd81f1dc971569a9b729c43f2f0e6436637
- Last updated
- 2024-01-04
SAVRN's Notes on race
RACE is machine-comprehension multiple choice drawn from Chinese middle- and high-school English examinations: 97,687 rows in the "all" configuration, 195,374 rows across the three configurations, 83.0 MB and 176.6k monthly downloads. The split table shows train at 87,866 rows, with test and validation at 4,934 and 4,887, so choose your configuration deliberately rather than training on the union by accident.
The license field reads "other", so confirm terms with the source before commercial use. Substantively, this is a benchmark set from the reading-comprehension literature, and it is standardized enough that contamination is a real concern: models pretrained at scale have seen it. Fine as a quick eval at 83.0 MB total; we would not build production training pipelines around it.
Files
11 files, 83.0 MB in total.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| all/test-00000-of-00001.parquet | Data | 2.1 MB | 442e6b5f22f5 |
| all/train-00000-of-00001.parquet | Data | 37.4 MB | d9986432dc7b |
| all/validation-00000-of-00001.parquet | Data | 2.0 MB | 2b3668ef8556 |
| high/test-00000-of-00001.parquet | Data | 1.7 MB | 97ef697f4c81 |
| high/train-00000-of-00001.parquet | Data | 30.4 MB | ea4766ae6849 |
| high/validation-00000-of-00001.parquet | Data | 1.7 MB | 9fc6a837dfa7 |
| middle/test-00000-of-00001.parquet | Data | 404.7 KB | 4ea3fb86ab6b |
| middle/train-00000-of-00001.parquet | Data | 7.0 MB | 33a9d28467c3 |
| middle/validation-00000-of-00001.parquet | Data | 406.9 KB | c8ede7668382 |
| README.md | Documentation | 11.0 KB | — |
| .gitattributes | Repository | 1.2 KB | — |
License and Download
- License
- other
- Access
- No access gate
Released by Eduard Hovy through its official repository on Hugging Face.